How AI Is Quietly Changing the Way We Shop

A few weeks ago, OpenAI quietly launched a new feature you might have missed. You can now ask ChatGPT to buy something, not just recommend it.

If you’re in the US, you can now type something like “Find me a gift for my sister under $50,” and suddenly, you’re looking at real Etsy listings, complete with photos, reviews, and an “instant checkout” button.

On the surface, it looks like just another integration. But underneath, this marks a much bigger shift.

ChatGPT is becoming a new kind of retail platform. And that raises some fascinating questions about visibility, design, and trust.

We’ve officially reached the point where you can shop without ever visiting a website — and that changes everything.

---

What exactly is ChatGPT’s Instant Checkout

ChatGPT has partnered with Stripe, Etsy and Shopify to allow users to buy items without leaving the chat window with their new Instant Checkout feature.

The key points:

• Launched: 9 September 2025, for US ChatGPT (Plus, Pro and Free) users

• Now: Can buy single-item purchases directly from US Etsy sellers

• Next: Over one million Shopify merchants coming soon, multi-item carts, and expand merchants and regions

• Paid ChatGPT subscribers can pay with their card on file, but other payment options are also available

• Merchants pay ‘a small fee’ on completed purchases

They’re also open-sourcing the Agentic Commerce Protocol so that other merchants and developers can begin building integrations, and apply to have their products available for purchase through ChatGPT.

It feels like they’re just starting to explore this space and I’m interested to see how this changes and develops as more people start to use it. But the thing I really want to understand is how does ChatGPT decide which products to recommend?

---

Inside the ranking engine: How AI decides what we buy

In their launch post, they state:

“ChatGPT shows the most relevant products from across the web. Product results are organic and unsponsored, ranked purely on relevance to the user.”

If this is true, this could be quite the game changer for small retailers in particular who are often pushed out of the market by whoever has the biggest online advertising budget.

When I asked ChatGPT about it, it was still intentionally vague, but explained in a little more detail what it’s using here:

• Relevance to the user’s prompt (“find me a ceramic mug under $30”).

• Availability — in stock and able to ship to the user’s location.

• Price and quality signals — presumably drawn from product metadata (price, reviews, seller rating).

• Merchant data completeness — structured feeds make it easier to surface listings accurately.

• Instant Checkout support — currently a soft preference, since those products can complete the purchase end-to-end in ChatGPT.

• No paid placement (for now) — OpenAI states listings are not “sponsored.”

We don’t yet know how those factors are weighted — but it’s safe to assume there’s a ranking logic quietly at work.

If the AI becomes the gatekeeper, how does that change the way people should design to make sure both customers and AI agents can find products online?

---

The UX you can’t see - Designing for an invisible storefront

A few months ago a friend was telling me about how they were finally able to find a jacket that fit them perfectly using ChatGPT. That positive experience established trust and now they regularly use it to find other things to shop for.

But back then they still had to link out to the retailers website and buy the product directly, which can lead to a different shopping experience.

When you click through to a retailer, you might find yourself searching to see if anyone else sells the same thing cheaper, reading more about it or buying additional items that were recommended to you on the retailers site.

If users start purchasing more directly via ChatGPT, they might never see the retailers website. This is quite the change from how most users shop currently.

When we think about online shopping, we usually picture the visual layer — product photos, layout, colours, copy. That’s where brands spend most of their energy: crafting experiences that feel tactile and beautiful.

But in this new world, retailers also need to consider what the experience is like for an AI agent. Some of this will be similar to how sites optimise for SEO, but it also feels like it needs to take that further.

How does a retailer help an AI agent understand this product listing is relevant, trustworthy and worth showing?

When agents do the discovery work, they rely on data, not design.

Clear titles, consistent product attributes, structured descriptions — those become the new storefront window for an AI agent.

It’s a strange inversion: the prettier your product page, the less it matters to an AI agent. What counts is whether your product data is coherent, standardised, and machine-readable.

If your description says “handmade ceramic mug,” but your tags are missing “dishwasher safe,” “capacity,” or “shipping time,” you’ve already lost the invisible UX battle.

---

Why we trust the bot (maybe too much)

The other thing I’ve been thinking about a bit recently is where people put their trust. When shopping online, most users check reviews before purchasing something, that helps increase their trust that this is the right product for them.

As we turn to AI powered shopping, there’s an expectation that the AI has done all the heavy lifting, read the reviews, analysed all the data and considered all the angles so whatever it recommends is perhaps better than what you would have found on your own.

But what if it hasn’t. We know that AI’s still hallucinate at high rates. When ChatGPT recommends a product, most people won’t stop to wonder how it decided. We’ll just assume it’s looked at every option, compared prices, filtered out the junk, and landed on the best choice.

Recently I used ChatGPT to help plan a holiday. I asked the AI to recommend some hotels, and hoped it would consider all the things I wanted, and find me the best deal saving me hours of time research. But what I actually found was that it often just promoted whichever hotels had the most articles written about them, quite often big name brands. Doing my own research I was able to find better options — which I could then put back into ChatGPT and it did a great job at helping me compare and make a decision. But if I had of assumed that it had considered every option already, I would have ended up staying somewhere that wasn’t actually what I wanted.

But I think the biggest risk here isn’t that AI makes bad choices — humans do that all the time.

It’s that we stop questioning how choices are made, or who benefits from them.

I’d love for ChatGPT to explain why it has recommended certain choices. That would help me decide if this is the right product for what I actually need. But I also still have concerns that it would just tell me what I want to hear and not perhaps the total truth.

Maybe the next evolution of AI powered e-commerce isn’t just smarter recommendations, but more transparent ones. Where the AI doesn’t just tell us what to buy, but why it thinks it’s right for us. That could be the real game changer here — and where the next big UX challenge begins.

Takeaway: In the age of agentic commerce, your next customer might not be human. The challenge now isn’t just making better products and websites — it’s designing data that speaks to the bots who might be the ones deciding what we see and what we buy.